Large-Scale CT Dataset with Complete Phase Coverage Advances Primary Liver Cancer AI Diagnostics

Take-Away Points

• ■ Major Focus: To develop a publicly available, multiphase three-dimensional (3D) contrast-enhanced CT (CECT) dataset of 278 patients with primary liver cancer, including hepatocellular carcinoma, intrahepatic cholangiocarcinoma, and combined hepatocellular-cholangiocarcinoma, featuring complete four-phase imaging (plain, arterial, venous, and delayed) and expert lesion annotations.

• ■ Key Results: The dataset includes over 50 000 manually annotated lesion sections with a median interobserver Dice coefficient of 0.76 and incorporates 83 nonliver cancer control cases, each with complete four-phase CECT scans.

• ■ Impact: This resource addresses critical gaps in publicly available liver cancer imaging data and provides multiphase, expertly annotated CECT images to support the development of accurate deep learning models for liver cancer classification and segmentation.

Primary liver cancer is a major global health burden, ranking as the sixth most common cancer worldwide and contributing substantially to cancer-related mortality. Accurate subtype classification, including hepatocellular carcinoma, intrahepatic cholangiocarcinoma, and combined hepatocellular-cholangiocarcinoma, is essential for guiding treatment decisions. However, available public datasets are often limited in subtype representation and lack complete multiphase imaging, impeding artificial intelligence (AI) development for diagnostic modeling.

To address this gap, Luo et al created a comprehensive 3D CECT dataset including 278 primary liver cancer cases with full four-phase imaging and clinician-verified liver and lesion masks. Rigorous validation was performed, including de-identification, conversion to Neuroimaging Informatics Technology Initiative format, and interobserver agreement analysis (median Dice coefficient, 0.76). Each case contains plain, arterial, venous, and delayed-phase images, with corresponding annotations. The dataset also includes 83 nonliver cancer controls, facilitating supervised model training.

This work establishes a fully annotated, multiphase CECT dataset to support development of robust AI models for liver cancer diagnosis and segmentation. While limited by its single-center design, this dataset provides a strong foundation for future research and multicenter validation across varied clinical settings and imaging protocols.

Highlighted Article

• Luo J, Wan X, Du J, et al. Comprehensive multi-phase 3D contrast-enhanced CT imaging for primary liver cancer. Sci Data 2025;12(1):768. doi: https://doi.org/10.1038/s41597-025-05125-2